From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1755325AbYDBIow (ORCPT ); Wed, 2 Apr 2008 04:44:52 -0400 Received: (majordomo@vger.kernel.org) by vger.kernel.org id S1751886AbYDBIon (ORCPT ); Wed, 2 Apr 2008 04:44:43 -0400 Received: from pentafluge.infradead.org ([213.146.154.40]:39224 "EHLO pentafluge.infradead.org" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1751901AbYDBIom (ORCPT ); Wed, 2 Apr 2008 04:44:42 -0400 Subject: Re: [PATCH 2/2] Customize sched domain via cpuset From: Peter Zijlstra To: Hidetoshi Seto Cc: linux-kernel@vger.kernel.org, Ingo Molnar , Paul Jackson In-Reply-To: <47F346A8.5000801@jp.fujitsu.com> References: <47F21C38.8090608@jp.fujitsu.com> <1207050712.8514.715.camel@twins> <47F346A8.5000801@jp.fujitsu.com> Content-Type: text/plain Date: Wed, 02 Apr 2008 10:44:31 +0200 Message-Id: <1207125871.8514.755.camel@twins> Mime-Version: 1.0 X-Mailer: Evolution 2.21.92 Content-Transfer-Encoding: 7bit Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org On Wed, 2008-04-02 at 17:41 +0900, Hidetoshi Seto wrote: > Peter Zijlstra wrote: > > On Tue, 2008-04-01 at 20:27 +0900, Hidetoshi Seto wrote: > >> The implementation is here. > >> > >> - Add 2 new cpuset files: > >> sched_wake_idle_far > >> sched_balance_newidle_far > >> > >> - Modify partition_sched_domains() and build_sched_domains() > >> to take flags parameter passed from cpuset. > >> > >> - Fill newidle_idx for node domains which currently unused but > >> might be required for sched_balance_newidle_far. > > > > Just to be clear; the same effect can be had by poking into: > > > > /proc/sys/kernel/sched_domain/$cpu/$domain/flags > > > > but this interface you now propose gives a more stable interface in that > > you'd have to re-do your setting after every cpuset change (admittedly > > those are rare, but I see how it could be a nuisance). > > And the sysctl entry "sched_domain" is not available unless SCHED_DEBUG. > > So it is common sense that this sysctl is not open to public yet, > and that the expected users are scheduler developers, Ingo and friends. Ah, right, totally forgot about that :-) > > Or do you actually add something that wasn't available through the > > initial domain interface? > > At this time I have no idea, but it would be possible if there are > something unreasonable on global system but acceptable on a part. > > In other words, we can invent other sched_* families which has special > effect that "default scheduler" never have. I was asking about this patch in specific, and the answer seems to be: no, we don't add anything that wasn't available. And yes, I see the possiblilties to extend this :-)